savings and income distribution

ANNALS OF ECONOMICS AND FINANCE 5, 245–270 (2004)

Savings and Income Distribution*

Hongyi Li

BA Faculty, Chinese University of Hong Kong, Hong Kong

and

Heng-fu Zou

The World Bank, Room MC2-611, 1818 H St. NW, Washington, D.C. 20433, USAGuanghua School of Management, Peking University, Beijing, 100871, China

Institute for Advanced Study, Wuhan University, Wuhan, 430072, China

While this paper emphasizes the analytical ambiguity of the relationship be-tween savings and income inequality, the empirical examination renders weaksupport for a negative association between them. However, this relationshipis not very robust. Subsamples of OECD countries and Asian countries showthat income inequality and the savings rate can be positively and significantlyassociated. c© 2004 Peking University Press

Key Words: Savings; Income distribution.JEL Classification Numbers: D3, E2, O5.

1. INTRODUCTION

Many recent studies have explored the relationship between income dis-tribution and economic growth (see, for example, Alesina and Rodrik, 1994,Persson and Tabellini, 1994), without examining the impact of income dis-tribution on savings. This is unfortunate because savings and investmentare the driving force of economic growth, and a detailed examination of therelationship between savings and income distribution should proceed be-fore any systematic study on the impact of income distribution on incomegrowth. Our current study intends to fill up this missing gap.

*Hongyi Li would like to acknowledge the financial support given by the Direct Grant,BA Faculty, of the Chinese University of Hong Kong.

Heng-fu Zou would like to acknowledge the financial support given by the NationalNatural Science Foundation of China (70225001).

2451529-7373/2004

Copyright c© 2004 by Peking University PressAll rights of reproduction in any form reserved.

246 HONGYI LI AND HENG-FU ZOU

The literature on savings behavior is enormous. For a detailed survey, seeAghevli et al. (1990). In general empirical evidence shows that countrieswith a high savings rate, such as Japan, South Korea, and Singapore aretypically associated with high growth rates. Kessler et al. (1993) discussthe savings behavior in 17 OECD countries. They find that a large partof the variation in savings behavior van be attributed to the variationacross countries. Masson et al. (1995) look at a broad set of possibledeterminants of savings, such as income growth, financial liberalization,interest rates, terms of trade, and apply them to savings data for a largenumber of industrial and developing countries. Edwards (1995) studies thedeterminants of savings for a panel of 36 countries and discusses the crosscountry differences in the saving ratios. His main results show that percapita income growth, political instability and financial development areimportant determinants of savings.

This paper traces the impact of income distribution on savings bothanalytically and empirically, thereby adding an important dimension toprevious studies on savings. This task is timely because income distributionhas received an enormous amount of attention recently, and has been mademuch easier using an improved data set on income distribution compiledby Deininger and Squire (1996). Previous empirical studies on incomedistribution have been hampered by data sets with very different definitionsof the Gini coefficients and very few observations over time and acrosscountries. Using this newly compiled and greatly expanded data set onincome distribution, we can minimize the methodological differences inthe definitions of Gini coefficients by selecting the Gini coefficients fromnational coverage household survey based on gross income, net income orexpenditure. Since we have explicit information on the definitions of Ginicoefficients, we are able to control the differences in measuring incomedistribution based on different criteria and maintain consistency in thedefinitions of Gini coefficients.

Using the new data set, we found by variance decomposition analysisthat income distribution is relatively stable within each country over time,but is significantly different across countries. Because savings rates alsovary significantly across countries, it is only natural to ask whether thisvariation has anything to do with the large variation in income distributionand other structural variables such as the initial income distribution, schoolattainment, financial development, and the civil liberty index.

Section 2 briefly discusses the patterns of savings and income distribu-tion. Section 3 presents an analytical model explaining the relationshipbetween savings and income distribution. Section 4 empirically examinethe relationship between savings and income distribution after controllingthe effects on savings of income growth, the interest rate, financial devel-opment, and the political system. Section 5 concludes.

SAVINGS AND INCOME DISTRIBUTION 247

2. PATTERNS OF SAVINGS BEHAVIOR AND INCOMEDISTRIBUTION

We first study the empirical patterns of savings behavior and incomedistribution in a panel of 49 industrial and developing countries. The sav-ings rate is defined in three ways based on available data. In particular,SR is the ratio of gross domestic savings (GDS) to gross domestic product(GDP), NSR is the ratio of gross national savings (GNS) to gross domesticproduct, and NSRP is the ratio of domestic private savings (GNPS) togross domestic product. For the sources and descriptions of the data, seethe Data Appendix.

The data for SR for most of the countries in our sample cover a periodfrom 1960 to 1992. For NSR and NSRP the coverage is from 1970 to1990. The income distribution data (Gini coefficients from 1947 to 1994)are taken from Deininger and Squire (1996). For most of the countries thenumber of valid observations of the Gini coefficient is usually small. Thesummary statistics of the savings rates and Gini coefficients are presentedin Table 1.1

TABLE 1.

Summary Statistics of Savings Rates and Gini Coefficients for 49 Countries

COUNTRY Obs. # MEAN STDE MAX MIN MAX-MIN COVERAGE

SR 1473 23.53 8.02 55.15 −3.29 58.54 1960 ∼ 92

NSR 1127 22.51 8.42 47.55 −3.86 51.41 1970 ∼ 92

NSRP 913 19.33 7.46 43.70 −8.79 52.49 1970 ∼ 92

GINI 556 36.09 9.03 61.88 18.44 43.44 1947 ∼ 94

For NSRP no data is available for BHS, BGR, CHN, CRI, JAM and PAN.

A direct visual inspection of the plot of the savings rate and the Ginicoefficient data reveals trends in the savings rate or the Gini coefficienttime series for some of the countries. Thus we estimate a simple lineartrend model to test the trends in the savings rate

Sit = αi + βit + uit (1)

where Sit (= SRit, NSRit, or NSRPit) is the savings rate; uit ∼ iid(0, σ2u).

Due to the differences in the definitions for the Gini coefficients, we usea least squares dummy variables (LSDV) regression instead

Giniit = φiDi + θiti +∑

kλkDctrl,k + ωit (2)

1For detailed analysis on the explanation of the variations of income inequality acrosscountries and over time, see Li, Squire, and Zou (1998). The summary statistics are alsoavailable for each country upon request.


where Giniit is the Gini coefficient, Di = 1 for country i/0 otherwise,and ω ∼ iid(0, σ2

ω). The dummy variables Dctrl,k (k = 1, 2, 3) controlfor the different definitions of the Gini coefficient in the original sources.The control dummies are defined as Dctrl,1 = 1 if gross income based/0otherwise; Dctrl,2 = 1 if net income based/0 otherwise; and Dctrl,3 = 1 ifhousehold based/0 otherwise. In both (1) and (2) i = 1, 2, . . . , N (numberof countries) and ti = 1, 2, . . . , Ti (number of years covered).2

Tables 2 and 3 present the trend estimation results. For the savingsrate, a large proportion of the countries have significant trends. For SR,21 countries have significant negative trends and 19 countries have sig-nificant positive trends; for NSR, 18 have significant negative trends and16 have significant positive trends; for NSRP, 11 have significant negativetrends and 14 have significant positive trends. For NSRP only 40 countriesare considered in the trend estimation due to data availability. Almostall the OECD countries in our sample have significant negative trends,compared to most Asian countries, or in general the developing countries,which exhibit significant positive trends in savings rates. Recent compara-tive studies of broad trends in savings behavior can be found in Shafer, etal. (1992), Maddison (1992), among other authors.

For the Gini coefficients, the number of countries that have significanttrends is smaller. See results in Table 3. Only half of the countries havesignificant trends, although the results are highly tentative because thetime series of the Gini coefficients have few and non-consecutive observa-tions. In the LSDV regression for the Gini coefficients, definition dummiesare included to count for the systematic differences in measuring incomedistribution. The general pattern of income distribution is relatively stablewithin a country and varies significantly across countries. The estimationresults indicate that Gini coefficients based on gross income are signifi-cantly higher (4 points on a scale of 1 to 100) than Gini coefficients basedon the other definitions such as net income or expenditure. Thus in ourempirical analysis in section 4, the definition-adjusted Gini coefficients areused instead.

3. THE ANALYTICAL AMBIGUITIES ON SAVINGS,INCOME DISTRIBUTION AND POLITICAL ECONOMY

Analytically the relationship between income distribution and savingshas been studied since the classical contributions by Lewis (1954), Kaldor(1957) and Pasinetti (1962). But in most of these models, the savings be-havior of different social groups and classes is assumed to be very different in

2Lagged values can be added in the trend regressions for savings rate and Gini coef-ficients to test for serial correlation. Because of missing data in the Gini series, we donot pursue this issue.


TABLE 2.

Trend Estimation for Savings Rates

SR NSR NSRP

COUNTRY NOB TREND t-VALUE NOB TREND t-VALUE NOB TREND t-VALUE

AUS 33 −0.19 −5.46 23 −0.21 −3.97 23 −0.21 −4.61

BHS 11 1.08 3.42 23 0.19 0.80

BGD 20 0.15 2.08 23 0.22 3.72 23 0.30 3.63

BEL 33 −0.07 −1.34 23 −0.32 −3.00 23 0.08 1.60

BRA 33 0.23 5.75 23 0.09 1.07 23 −0.08 −0.76

BGR 12 −0.76 −2.69 23 −0.88 −7.58

CAN 33 −0.10 −2.91 23 −0.35 −6.59 23 0.03 0.50

CHL 33 0.30 3.20 23 0.82 4.49 23 1.00 8.32

CHN 16 0.69 10.26 23 0.01 0.07

COL 33 0.20 5.13 23 0.06 0.59 23 −0.10 −1.06

CRI 33 0.42 7.83 23 0.59 5.03

CIV 32 −0.36 −4.36 23 −2.25 −14.97 23 −1.50 −6.56

CSK 12 −0.66 −2.89 23 −0.15 −3.53 23 −0.20 −9.26

FIN 33 −0.09 −2.20 23 −0.88 −8.78 23 0.22 1.89

FRA 33 −0.23 −7.59 23 −0.32 −6.35 23 −0.13 −3.64

DFA 33 −0.16 −3.99 23 −0.12 −1.82 23 0.10 2.40

HND 33 0.05 0.78 23 0.13 0.97

HKG 32 0.40 7.66 23 0.73 7.76 12 0.71 3.12

HUN 23 −0.46 −5.56 23 −0.81 −7.02

IND 33 0.32 13.36 23 0.13 2.18 23 0.21 3.48

IDN 33 1.00 8.60 23 0.85 9.25

IRN 19 −0.76 −2.96 23 −0.55 −2.71 23 −0.52 −2.75

ITA 33 −0.31 −15.73 23 −0.38 −11.09 23 −0.18 −3.97

JAM 33 −0.30 −2.55 23 0.29 1.99

JPN 33 −0.13 −2.88 23 −0.19 −2.49 23 −0.32 −13.19

KOR 33 1.13 25.67 23 0.99 9.38 23 0.78 8.04

MYS 33 0.35 6.39 23 0.51 4.64 23 0.65 5.14

MEX 33 0.23 4.58 23 0.26 2.27 23 0.22 1.66

NLD 33 −0.18 −5.00 23 −0.14 −1.81 23 0.24 4.25

NZL 33 −0.11 −2.89 23 −0.64 −4.44 23 −0.41 −2.17

NOR 33 0.03 0.75 23 −0.16 −1.92 23 −0.02 −0.25

PAK 33 0.11 2.37 23 0.45 4.14 23 0.62 6.64

PAN 33 0.14 1.78 23 −0.64 −6.57

PER 33 −0.78 −5.89 23 −0.22 −1.45 23 0.22 1.26

PHL 33 0.03 0.61 23 −0.02 −0.33 23 −0.17 −2.48

POL 13 0.32 0.54 23 −0.13 −0.84 23 −0.03 −0.22

PRT 33 0.03 0.42 23 0.90 8.33 23 0.92 8.95

SGP 33 1.56 19.48 23 1.40 8.86 23 1.31 9.34

ESP 33 −0.16 −5.81 23 −0.09 −1.92 23 −0.12 −2.45

LKA 33 0.05 1.09 23 0.47 4.83 23 0.45 4.37

SWE 33 −0.23 −6.88 23 −0.30 −4.46 23 0.09 1.26

TWN 33 0.56 6.38 23 0.29 2.58 16 0.21 0.71

THA 33 0.45 9.10 23 0.60 5.10 23 0.36 4.49

TTO 33 −0.17 −1.18 23 −1.47 −9.02 23 0.12 0.95

TUN 32 0.23 3.63 23 0.06 0.53 23 0.77 6.30

GBR 33 −0.08 −2.92 23 −0.35 −9.31 23 −0.10 −1.51

USA 33 −0.15 −6.44 23 −0.21 −4.26 23 −0.09 −2.19

VEN 33 −0.65 −5.71 23 −0.36 −1.79 23 0.11 0.64

YUG 30 0.34 3.22 23 −0.02 −0.22 23 −0.04 −0.40

1. For SR, 21 countries have significant negative trends, 19 countries have significant positive trends. (Atotal of 40 countries have significant trends.) A 5% t-test is used. Same below.2. For NSR, 18 countries have significant negative trends, 16 countries have significant positive trends.(A total of 34 countries have significant trends.)3. For NSRP, 11 countries have significant negative trends, 14 countries have significant positive trends. (Atotal of 25 countries have significant trends.) For HND, HUN and IDN, since the number of observationsis small, the results are not reported.


TABLE 3.

LSDV Estimation (Gini Coefficients )

Control Dummy Estimate t-VALUE

Gross 4.00 4.29

Net 1.60 1.47

HHtype 0.41 0.71

Constant t-VALUE

Country Obs Estimate t-VALUE Estimate t-VALUE

AUS 10 31.09 26.26 0.02 0.15

BHS 11 42.23 35.80 −0.30 −3.78

BGD 9 31.43 24.60 0.13 1.37

BEL 9 31.93 26.83 −0.44 −4.45

BRA 14 53.57 52.91 −0.08 −0.84

BGR 27 19.21 19.74 0.15 3.13

CAN 23 26.63 24.90 −0.03 −0.86

CHL 15 47.86 43.57 0.34 4.04

CHN 12 23.57 15.20 0.79 4.72

COL 7 46.81 38.01 0.05 0.50

CRI 8 41.81 35.44 −0.23 −2.51

CIV 5 39.11 9.69 0.06 0.11

CSK 10 19.29 14.60 −0.18 −2.47

FIN 9 28.47 22.46 0.23 1.72

FRA 7 34.88 23.05 −0.46 −5.40

DEU 8 32.76 26.70 0.36 4.06

HND 6 52.96 37.02 −0.36 −3.59

HKG 10 38.45 31.28 −0.14 −2.20

HUN 8 23.00 18.18 0.12 1.60

IND 29 31.26 59.22 −0.10 −3.29

IDN 7 34.99 39.59 −0.30 −1.79

IRN 6 42.82 28.50 −0.07 −0.39

ITA 15 33.45 27.02 −0.37 −3.44

JAM 5 44.49 45.74 −0.21 −2.64

JPN 22 29.77 26.25 −0.07 −1.18

KOR 10 30.78 24.82 0.14 1.61

MYS 6 45.75 35.83 −0.16 −1.21

MEX 9 43.93 32.87 −0.33 −5.49

NLD 10 25.13 15.51 0.33 1.83

NZL 11 28.29 24.43 0.30 2.40

NOR 9 31.91 25.54 −0.20 −2.54

PAK 12 30.90 29.11 −0.09 −1.35

PAN 4 48.28 34.76 −0.01 −0.06

PER 5 46.13 37.67 −0.55 −4.46

PHL 7 38.99 32.47 −0.03 −0.50

POL 8 23.54 14.84 0.02 0.11

PRT 4 33.79 22.39 −0.22 −1.56

SGP 6 35.66 26.98 0.02 0.09

ESP 8 26.83 27.98 −0.21 −2.08

LKA 9 35.25 35.27 −0.27 −3.96

SWE 15 28.99 25.54 −0.10 −1.00

TWN 26 28.01 24.15 −0.02 −0.44

THA 8 40.90 33.55 0.31 4.23

TTO 4 40.53 22.96 −0.14 −1.19

TUN 5 42.61 45.16 −0.04 −0.40

GBR 31 25.11 21.64 0.18 4.38

USA 45 31.53 30.03 0.06 2.58

VEN 8 38.08 31.35 0.25 1.70

YUG 4 30.20 21.74 0.22 2.09

14 countries have significant negative trend, 11 countries have significant positivetrend. (A total of 25 countries has significant trends.) A 5% t-test is used.


an ad hoc manner. For example, Lewis (1954) assumes that entrepreneurssave a larger fraction of their profit income than the other groups in theeconomy; Kaldor (1957) takes the savings rate of the working class to bezero. Thus income inequalities can generate high savings rate if the richhave a larger share of income in an economy. Until now this line of researchhas not been seriously developed or tested empirically.

In this section, we develop an analytical model to illustrate how incomedistribution affects savings through both the life-cycle permanent-incomeargument and the political-economy mechanism. We will see that, unlikethe ad hoc models, it is very difficult to derive a definite relationship be-tween income distribution and savings from sound microfoundations.

Consider a typical overlapping-generations model. Member h (h = 1, . . . ,H(t)) born at time t has a utility function defined on his/her consumptionwhen young, ch

t (t), his/her consumption when old, cht (t + 1), public con-

sumption when young, g(t), and public consumption when old, g(t + 1)

uht (ch

t (t), cht (t + 1), g(t), g(t + 1)). (3)

His/her budget constraint is given by

cht (t) + sh

t (t) = ωht (t) (4)

cht (t + 1) = (1− τ(t))(1 + r)sh

t (t) (5)

where sht (t) is his/her savings when he/she is young, ωh

t (t) is his/her initialincome, τ(t) is the rate of capital income taxation on generation t, r is theexogenously given interest rate on savings. H(t) is the total population ofgeneration t. Public consumption is financed by the government throughcapital income taxation with g(t) levied on the time t old and g(t+1) leviedon the time (t + 1) old:

g(t) =H(t−1)∑

h=1

τ(t− 1)(1 + r)sht−1(t− 1) (6)

g(t + 1) =H(t)∑h=1

τ(t)(1 + r)sht (t) (7)

where H(t − 1) and H(t) are the numbers of people of generations t and(t + 1), respectively.

In this setup, agent h has a savings decision to make in order to smoothconsumption over time. The savings by agent h will be determined byhis/her preferences, initial income distribution, capital income taxation,


interest rate, and government spending. Without simply assuming awaythe decision of public spending, we follow the political economy argumentsin Alesina and Rodrik (1994), Persson and Tabellini (1994), and Perroti(1993), and propose a voting mechanism on public goods: each agent max-imizes his/her utility defined on both private and public goods by choosingthe most desired rate of capital income tax. Of course, his/her choices willbe in turn determined by various factors mentioned above, in particular,initial income. Then the society’s collective choice through the majorityvoting will be determined by its income distribution and other factors.

To see this more clearly, let us first solve agent h’s consumption and sav-ings problem. Given the choices of g(t) and g(t+1), agent h’s optimizationyields the following first-order condition:

∂ uht (ωh

t (t)− sht (t), (1− τ(t))(1 + r)sh

t (t), g(t), g(t + 1))∂ ch

t (t)

=∂ uh

t (ωht (t)− sh

t (t), (1− τ(t))(1 + r)sht (t), g(t), g(t + 1))

∂ cht (t + 1)

(1− τ(t))(1 + r).

From this condition, we have

sht (t) = sh

t (ωht (t), (1− τ(t))(1 + r), g(t), g(t + 1)). (8)

By summing savings by the time t young, we have the total savings attime t denoted as s(t):

s(t) =∑

sht (t) =

∑sh

t (ωht (t), (1− τ(t))(1 + r), g(t), g(t + 1)). (9)

At this general level, we have the following general result: If, in bothperiods, consumption is normal, then

∂sht (t)

∂ωht

> 0.

But to link savings to income distribution it is necessary to know the signof ∂2sh

t (t)

∂ωht (t)2

. If ∂2sht (t)

∂ωht (t)2

< 0, then a more equal distribution leads to highersavings because the marginal savings rate from income is decreasing; onthe other hand, when ∂2sh

t (t)

∂ωht (t)2

> 0, a more unequal distribution will leadto more savings because the marginal savings are increasing in income. Ofcourse, if ∂2sh

t (t)

∂ωht (t)2

= 0, then it does not matter how income is distributedamong the members of the society.

Next, we turn to the political-economy argument for the determinationof the tax rate and public service provision. We assume that every member


of this society maximizes his/her utility by choosing the most preferred taxrate

max[τ(t)]

uht (ωh

t (t)− sht (t), (1− τ(t))(1 + r)sh

t (t), g(t), g(t + 1))

subject to (6), (7) and (8). The society will determine the tax rate τ(t)by the majority rule. This choice of the tax rate is necessarily determinedby each voter’s income share among his/her generation, or more generally,τ(t) is a function of income distribution:

τ(t) = τ({ωht (t)}H(t)

h=1 ). (10)

As in Alesina and Rodrik (1994), when income distribution for this econ-omy is more equal, the median voter’s income share is higher. Since thetax rate is linked to the median voter’s income share, therefore, incomedistribution affects the chosen tax rate. It should be noted that with ourpreference structure and government budget constraint the time t youngdetermine g(t+1) and τ(t) while the time t old determine g(t) and τ(t−1).

Since the interest-rate effect on savings is ambiguous as a result of in-come and substitution effects, we cannot predict the effect of capital incometaxation on savings at this general level. From this observation, it is imme-diately apparent that a high capital income tax may increase or decreasesavings by agent h when he/she is young. This is to say, a higher or alower tax rate as a result of income distribution has an ambiguous impacton savings by the young.

At any time t, the time t young save while the time t old just consumewhat they have saved plus the after-tax return. Thus, the aggregate savingsrate at time t, denoted as S(t), is a complicated function of income distri-bution ({ωh

t−1(t − 1)}H(t−1)h=1 , {ωh

t (t)}H(t−1)h=1 ), population (H(t − 1), H(t)),

interest rate (r), and capital income tax (τ({ωht (t)}H(t)

h=1 )) and other factors:

S(t) ≡H(t)∑h=1

sht (ωh

t (t), (1 + r)(1− τ({ωht (t)}H(t)

h=1 )))

H(t)∑h=1

ωht (t) + (1 + r)

H(t−1)∑h=1

sht−1(ω

ht−1(t− 1), (1 + r)(1− τ({ωh

t−1(t− 1)}H(t−1)h=1 )))

.

(11)

In this definition, we see clearly that both intergenerational and in-tragenerational income distribution can affect the aggregate savings ratenot only through pure economic reasoning, but also through the political-


economy mechanism of capital income taxation τ({ωht−1(t−1)}H(t−1)

h=1 ) andτ({ωh

t (t)}H(t)h=1 ).

Equation (11) provides the fundamental analytical result for our empir-ical examination of the impact of income distribution on savings acrosscountries. To make it empirically operational, we take the following ap-proximation:

S(t) = S(Gini(t),∆(t), r(t), λ(t), p(t)) (12)

where Gini(t) is the Gini coefficient of household income, as a measure of

income distribution; ∆ is the income growth rate( =∑

ωht (t)−

∑ωh

t−1(t−1)∑ωh

t−1(t−1)−

1); r(t) is the interest rate; λ(t) is a measure of financial development; andp(t) is a measure of political systems.

It should be noted that we have incorporated λ(t) and p(t) to the savings-rate function. This is important because savings are mobilized throughthe financial structure and taxation is determined and implemented in aparticular political system. In different political regimes, even the majorityrule on taxation will function differently; see Edwards (1995) for morearguments on this point.

As argued earlier, in general it is not possible to make definite theoreticalpredictions on the effects of income distribution on savings. To make thispoint clear, we offer two examples that shed light on our understanding ofthe relationship between savings and income distribution.

Example 1: Savings are independent of income distributionAll agents (h = 1, . . . ,H(t)) have an identical preference, and agent h

chooses consumption to maximize

α1 ln cht (t) + α2 ln ch

t (t + 1) + β1 ln g(t) + β2 ln g(t + 1)

subject to the budget constraint in (4) and (5) and with given g(t) andg(t+1). Since the preference is homothetic, it is well-known that agent h’ssavings are linear in his/her income. Therefore, the economy’s aggregatesavings rate is independent of income distribution.

To determine the rate of capital income taxation, we assume a major-ity rule. In this case, each agent h maximizes his/her utility by choosinghis/her most preferred tax rate. It is also well-known that, with a homo-thetic preference structure, all voters will want the same tax rate (Lovell,1975). Thus the optimal choice of income tax rate is also independent ofincome distribution.

Example 2: Savings depend on income distribution indirectlythrough income taxation

In example 1, let the common preference be changed to:

α1 ln cht (t) + α2c

ht (t + 1) + β1 ln g(t) + β2g(t + 1).


Then the savings for agent h become

sht (t) = ωh

t (t)− α1

α2(1 + r)(1− τ).

Now, each individual’s savings are still linear in his/her income. For agiven tax rate, the savings rate will be independent of income distributionamong the H(t) members of the time t generation because

S(t) =

∑ωh

t (t)− α1α2(1+r)(1−τ)H(t)∑

ωht (t) + (1 + r)

∑ωh

t−1(t− 1)− α1α2(1−τ)H(t− 1)

.

But unlike example 1, each individual’s savings are decreasing in thecapital income tax since

∂ sht (t)

∂ τ= − α1

α2(1 + r)(1− τ)2< 0.

To determine τ by the society, we first examine each individual’s pre-ferred capital income tax. With the required substitution of the savingsfunction in the budget constraint and the revenue constraint of the govern-ment in the utility function, we have the following first-order condition foragent h’s optimal choice of τh:

α1

(1− τh)− α2(1 + r)ωh

t (t) + β2

∑H(t)

j=1ωj

t (t)− β2α1H(t)

α2(1 + r)(1− τh)2= 0.

Let us define W (t) as the total income of the time t young as:

W (t) =H(t)∑j=1

ωjt (t)

and σht (t) as the income share of agent h of generation t:

σht (t) =

ωht (t)∑ H(t)

j=1 ωjt (t)

.

Then, we can rewrite the first-order condition as

α1

(1− τh)W (t)−α2(1+r)σh

t (t)+β2−β2α1

α2(1 + r)(1− τh)2W (t)/H(t)= 0.

To find out how agent h’s choice of τh changes with respect to his/herincome share σh

t (t) we differentiate the equation above:

dτh

dσht (t)

=α2

2(1 + r)2(1− τh)3W (t)α1α2(1 + r)(1− τh)− 2β2α1H(t)

.


Therefore, if α1α2(1 + r)(1− τh)− 2β2α1H(t) > 0, then dτh

dσht (t)

> 0; and

if α1α2(1 + r)(1 − τh) − 2β2α1H(t) < 0, then dτh

dσht (t)

< 0. With majorityvoting on income taxation, the tax rate will be determined by the medianvoter’s preferred choice. Let τm denote the optimal tax rate chosen by themedian voter and σm be the median voter’s income share. Then the taxrate chosen by the society is implicitly determined by the median vote’sfirst-order condition:

α1

(1− τm)W (t)−α2(1+r)σm(t)+β2−β2

α1

α2(1 + r)(1− τm)2W (t)/H(t)= 0.

Again, note that, as in Alesina and Rodrik (1994), more equal incomedistribution implies a higher income share for the median voter. In thiscase, income distribution affects the choice of the optimal tax rate of themedian voter and the savings of the whole economy. If α1α2(1 + r)(1 −τm) − 2β2α1H(t) < 0, and if the income distribution is more equal (i.e.,a large income share for the median voter), the median voter will vote fora smaller tax rate. Then everyone will save more. Hence a more equalincome distribution leads to a higher savings rate. On the other hand, ifα1α2(1+ r)(1− τm)− 2β2α1H(t) > 0, a more equal income distribution (alarger income share for the median voter) will lead to more income taxationand less savings.

4. EMPIRICAL RESULTS

Our theory provides us with a framework identifying the main determi-nants of savings, but predicts an ambiguous effect of income distributionon savings. In the examples we studied, more equal income distributioncan increase aggregate savings, decrease aggregate savings, or not affectaggregate savings. Thus it is important to find out whether this ambiguitybetween savings and income distribution still holds empirically. For thispurpose, we estimate a linear regression model based on (12):

Sit = αi+β1Giniit+β2GGRWit+β3Rit+β4FDPit+β5POLIit+εit (13)

for i = 1, 2, . . . , N and t = 1, 2, . . . , T , where N is the number of coun-tries and T is the time period. In (13), the dependent variable Sit (=SRit, NSRit or NSRPit)3 is the savings rate; εit ∼ iid(0, σ2

ε).4 The in-dependent variables are: Giniit (Gini coefficients adjusted for differences

3The savings rate s is limited to the range [0, 1] (where negative savings is not consid-ered). Simple least squares is inappropriate. A transformation of ln(s/(1 − s)) is usedin the regression analysis.

4Since the Gini coefficients data for individual countries are in general time series withmany missing observations, we did not use various specification tests for serial correlation


in definitions), GGRWit (per capita real GDP growth), Rit (real inter-est rate), FDPit (financial development or financial depth measurement,defined as M2/GDP , where M2 is money (M1) plus quasi-money), andPOLIit (measure of political freedom, ranked from 1 to 7, with 1 for themost politically free countries and 7 the least politically free ones). For adescription of the variables and their sources, see the Data Appendix.

The pooled regression (when assuming α1 = α2 = · · · = αN ) and fixed-effects model are estimated. The fixed-effects model allows for countryspecific effects. Empirically, the evidence found in the literature regardingvariations across countries in savings behavior suggests both homogeneityand heterogeneity. Carroll et al. (1994) find no difference in savings be-havior based on a study of Canada immigrants. Kessler, et al. (1993), onthe other hand, argue that, even within the 17 OECD countries, there isno evidence of homogeneous savings behavior.

In our pooled regression, the OECD country dummy (ODM ) and theAsian country dummy (ADM ) are included to account for the regionaleffects discussed in the general patterns of savings behavior. For the fixed-effects model, the measure of political systems (averaged over time) is notincluded as a regressor because it has no time variation. We also repeat theestimations for subsamples of the OECD countries and the Asian countries.The estimation results are summarized in Tables 4, 5, and 6 where thedependent variable is SR, NSR or NSRP, respectively. Next, we discussthese results in details.

4.1. The relationship between gross domestic savings rate (SR)and income distribution4.1.1. Complete sample estimation

Table 4 summarizes the empirical results regarding regression (13) for thecomplete sample when the dependent variable is SR - the gross domesticsavings rate. Due to the lack of data on the other variables, some of the 49countries are not included in the estimation. For the complete sample weare left with 41 countries with 311 observations. For the pooled regression,different specifications (cases (1) to (4)) are used to test the stability of theregression coefficients. Case (5) gives the results from the estimation of thefixed-effects model. For regression results involving the fixed-effects spec-ification, there are indications that the fixed-effects hypothesis cannot berejected. However, given the fact that the current panel data is highly un-balanced (for some countries there are only a few observations), regressionresults based on the fixed-effects model should be explained with caution.

and causality, or lagged variables as instruments to account for possible endogeneity. Seesimilar work by Li and Zou (1998), Li, Xie, and Zou (2000), and Li, Xu, and Zou (2000).


In all cases, including case (5) for the fixed-effects model, the estimatesof the Gini coefficients are not significant. However, in four cases the signsare negative, suggesting that high inequality leads to low savings rate.

The coefficients on GDP growth are all positive and strongly significant.The positive impact of GDP growth on savings has also been found inMasson et al. (1995) for a panel of 64 developing countries in a fixed-effectsmodel. However, for the 21 industrial countries in their study, a negative(not significant) coefficient is found. It is different from our results forthe OECD countries, which will be discussed in the next subsection. Forfinancial depth, the coefficients are all positive and significant in all thepooled regressions (1) to (4). Edwards (1995) obtains similar results forprivate savings. In case (5), which allows for country specific effects, thecoefficient of financial depth becomes negative although not significant.

The signs of the coefficients on the real interest rate could be either posi-tive or negative, depending on whether the income effect or the substitutioneffect dominants. We find that the coefficients are negative in cases (3) and(5), and positive in (1), (2) and (4). In all cases, the coefficients are notsignificant.

Please note that in case (1), political freedom has a significant and pos-itive coefficient, which suggests that, in general, politically less free coun-tries have higher savings rates. When we add the OECD and Asian countrydummies, the coefficient of political freedom becomes insignificant. How-ever, the coefficient of the OECD dummy estimate is negative and sig-nificant, which essentially leads to the same conclusion since the OECDcountries are politically more free than other countries. The explainingpower clearly shifts to the OECD and Asian country dummies.

4.1.2. The OECD countries

We consider two special cases. The OECD country and the Asian coun-try subsamples. As we know from the summary statistics, most of theOECD countries have negative trends in savings rates, while almost all theAsian countries have positive trends. This is also one of the stylized factsoften documented in the savings literature. For the OECD country sample,the regressions are based on 15 countries with 180 observations. The GDPgrowth has positive and significant estimates, which are similar to the com-plete sample. The financial depth has positive and significant coefficientsin the pooled regressions (cases (6) and (7)), while they are negative andsignificant in case (8), the fixed-effects model.

TA

BLE

4.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Gin

i

Dep

enden

tV

aria

ble:

SR

(Gro

ssD

om

esticSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−165.0

6−

128.6

6−

162.9

5−

128.7

7-

−170.6

2−

168.0

5-

−251.6

2−237.9

6-

(−11.4

1)

(−7.5

5)

(−11.1

8)

(−8.1

0)

(−11.0

3)(−

11.1

7)

(−4.0

6)

(−4.3

3)

GIN

I−

0.1

8−

0.6

70.1

1−

0.6

7−

0.2

50.5

90.5

90.4

30.8

80.7

70.0

5

(−0.4

6)

(−1.7

0)

(0.3

1)

(−1.7

1)

(−0.5

2)

(1.0

7)

(1.0

6)

(1.2

7)

(0.5

8)

(0.5

2)

(0.0

5)

GG

RW

3.4

33.2

84.1

03.2

82.3

83.9

23.9

62.3

25.6

55.9

31.3

1

(4.4

8)

(4.3

1)

(5.6

3)

(4.3

8)

(6.7

2)

(5.2

2)

(5.2

9)

(7.7

0)

(2.6

6)

(2.9

3)

(1.5

3)

R0.0

10.0

3−

0.0

10.0

3−

0.0

60.6

30.5

9−

0.8

10.6

00.5

30.1

5

(0.1

3)

(0.2

7)

(−0.0

5)

(0.2

7)

(−1.0

7)

(2.0

0)

(1.9

0)

(−5.4

0)

(0.7

0)

(0.6

4)

(0.4

1)

FD

P0.5

30.6

20.4

70.6

2−

0.1

00.2

20.2

5−

0.2

81.2

01.1

6−

0.0

3

(4.8

7)

(5.2

9)

(4.4

1)

(5.3

9)

(−1.0

8)

(2.0

0)

(2.2

6)

(−4.0

8)

(4.4

6)

(4.5

5)

(−0.1

8)

PO

LI

4.3

0−

0.0

43.0

12.3

7

(2.5

4)

(−0.0

2)

(0.7

3)

(0.4

8)

OD

M−

2.2

9−

2.2

9

(−3.8

9)

(−4.4

3)

AD

M−

0.4

9−

0.4

9

(−0.6

2)

(−0.6

5)

NO

C41

41

41

41

41

15

15

15

14

14

14

NO

B311

311

311

311

311

180

180

180

89

89

89

AD

J-R

SQ

0.1

70.2

10.1

60.2

10.8

60.2

10.2

20.8

90.3

10.3

10.9

2

Reg

ressions

(1)

-(4

),(6

),(7

),(9

)and

(10)

are

pooled

regressio

ns;

(5),

(8)

and

(11)

are

fixed

-effects

model

regressio

ns

(coeffi

cients

for

dum

my

varia

bles

not

reported

).T

he

t-valu

esare

reported

inparen

theses.

Sam

efo

rTables

5to

9.


The coefficients for Gini remain insignificant and positive ranging from0.43 to 0.59. For the OECD countries the political freedom index clearlyvaries less, which could be the main reason for an insignificant but positivecoefficient. The real interest rate has positive coefficients in the pooledregressions in cases (6) and (7). In the fixed-effects model, the coefficientis negative and significant.

4.1.3. The Asian countries

For the Asian countries (cases (9) to (11)), the coefficients for Gini rangefrom 0.05 to 0.88. Again, none of the coefficients for Gini is significant.GDP growth is positive and significant in all three cases. Financial depthis positive in all three cases but not significant in case (11). The coefficientsof real interest rate are positive, though insignificant.

In this exercise, one result comes out quite strong. GDP growth is highlycorrelated to the savings rate in all the three samples with different modelspecifications. In particular, we find that the relationship between thesavings rate and the Gini coefficient is weak and insignificant.

4.2. The relationship between gross national savings rate (NSR)and income distribution

The parallel results using NSR as the dependent variable are reported inTable 5. Because the coverage of NSR is from 1970 to 1992, the number ofobservations is smaller than the corresponding cases in Table 4. Other dif-ferences between SR and NSR can also be found in the summary statisticsreported in Table 1.5

Broadly speaking, the coefficient estimates show a similar pattern whencompared with the results in Tables 4 and 5. We highlight some of themajor differences and similarities here. One major difference is that, for thecomplete sample, Gini has all negative coefficients and is strongly significantin cases (1) to (4). For the OECD country sample, the coefficients in cases(6) and (7) are negative, although still insignificant.

5The major differences between SR and NSR are the time coverage and methodologyin calculating the savings ratios. Note that NSR (gross domestic savings divided byGNP, 1970-1992) is an International Monetary Fund (IMF) definition, while SR (1960-1992) is calculated as gross domestic savings divided by GDP using data from the WorldBank.

TA

BLE

5.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Gin

i

Dep

enden

tV

aria

ble:

NSR

(Gro

ssN

atio

nalSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−141.6

1−

106.5

3−

137.0

0−

108.9

1-

−154.1

3−146.8

8-

−200.6

0−184.3

0-

(−10.7

3)

(−7.0

4)

(−10.2

9)

(−7.7

9)

(−7.6

7)

(−7.4

7)

(−5.8

6)

(−6.0

7)

GIN

I−

1.4

3−

1.7

9−

1.1

6−

1.8

0−

0.2

9−

1.0

4−

1.0

60.5

20.2

00.0

72.2

5

(−4.2

4)

(−5.1

9)

(−3.5

2)

(−5.2

3)

(−0.3

7)

(−1.5

0)

(−1.5

3)

(0.9

8)

(0.2

4)

(0.0

8)

(2.1

7)

GG

RW

4.9

04.5

35.6

44.4

73.9

75.1

95.3

42.9

03.5

83.9

60.8

6

(7.2

6)

(6.8

4)

(8.8

1)

(6.8

9)

(7.1

8)

(5.3

5)

(5.5

0)

(5.8

8)

(3.0

2)

(3.5

1)

(1.0

8)

R0.0

50.0

60.0

30.0

6−

0.0

5−

0.2

2−

0.3

3−

1.4

20.2

70.2

10.1

6

(0.5

2)

(0.6

6)

(0.3

4)

(0.6

8)

(−0.5

7)

(−0.5

6)

(−0.8

5)

(−6.0

8)

(0.5

7)

(0.4

4)

(0.4

9)

FD

P0.6

10.6

10.5

40.6

20.0

50.5

50.6

0−

0.3

51.0

10.9

60.2

6

(6.3

7)

(5.9

2)

(5.7

5)

(6.1

2)

(0.3

0)

(3.9

4)

(4.4

3)

(−3.2

4)

(6.8

1)

(6.8

4)

(1.6

9)

PO

LI

4.6

2−

0.7

97.8

62.8

2

(3.0

7)

(−0.4

2)

(1.5

6)

(1.0

3)

OD

M−

2.0

9−

1.9

8

(−3.8

2)

(−4.1

2)

AD

M1.0

10.9

3

(1.4

7)

(1.4

1)

NO

C41

41

41

41

41

15

15

15

14

14

14

NO

B285

285

285

285

285

155

155

155

87

87

87

AD

J-R

SQ

0.3

70.4

10.3

50.4

20.6

80.2

60.2

50.8

30.4

90.4

90.8

5


With respect to the other variables, the pattern is basically consistent.The coefficients of GDP growth are all positive and significant except incase (11), where it is less significant. Real interest rate, again, has negativeand significant coefficients in case (8) for the OECD sample, but is notsignificant in any other cases.

4.3. The relationship between domestic private savings rate(NSRP) and income distribution

Finally, we report in Table 6 the estimation results with respect to thegross private savings rate (NSRP). The coverage of private savings rate isthe same as that of the gross national savings rate. Note that our the-oretical model is more relevant to the private savings rate. Empirically,income distribution data is based on national-level household-income sur-veys. Perhaps it better describes the relationship between savings andincome distribution when we use the domestic private savings ratio.

For the complete sample, the Gini has negative and significant coefficientsin cases (1) to (4), which indicates that higher inequality leads to lowerprivate savings. When we allow for country-specific effects in case (5) thecoefficient is positive, although insignificant. For the OECD sample, thecoefficients are all positive but insignificant. In case (11) for the Asiancountry sample, the coefficient is positive and significant when countryspecific effects are allowed.

Compared to Tables 4 and 5, GDP growth continues to have a strong pos-itive impact on private savings rate; except for the Asian country sample,the coefficients are positive but insignificant. Financial depth has positiveand significant coefficients except in cases of the fixed-effects model. Forthe OECD country sample, financial depth is negative and significant whenwe allow for country-specific effects.

The real interest rate is insignificant in most of the cases except in case(8) for the OECD sample, where it is negative and significant. This resultis the same for all three measures of savings rate. For the OECD countrysample, a negative and significant relationship between savings rate andreal interest rate is identified when we allow for country-specific effects.

TA

BLE

6.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Gin

i

Dep

enden

tV

aria

ble:

NSR

P(G

ross

Priva

teSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−118.4

4−

80.5

6−

118.6

3−

90.3

0-

−209.8

1−

204.8

9-

−175.8

4−186.0

1-

(−7.5

3)

(−4.5

9)

(−7.5

6)

(−5.2

3)

(−10.9

7)(−

11.0

0)

(−5.7

7)

(−6.1

2)

GIN

I−

2.0

3−

2.3

2−

1.9

8−

2.6

20.3

90.7

00.6

80.6

80.0

3−

0.1

73.9

0

(−4.7

5)

(−5.3

1)

(−5.0

7)

(−6.1

9)

(0.4

2)

(1.0

6)

(1.0

3)

(1.2

2)

(0.0

4)

(−0.2

1)

(2.7

4)

GG

RW

2.6

72.3

52.7

52.0

62.0

32.8

12.9

11.0

71.1

90.7

40.2

4

(3.1

1)

(2.7

9)

(3.4

0)

(2.4

5)

(3.0

9)

(3.0

4)

(3.1

7)

(2.0

6)

(1.0

8)

(0.6

9)

(0.2

6)

R0.1

50.1

50.1

50.1

7−

0.0

30.2

60.1

8−

0.7

6−

0.2

8−

0.1

00.3

6

(1.4

4)

(1.4

5)

(1.4

2)

(1.6

8)(−

0.3

2)

(0.7

0)

(0.5

0)

(−3.0

9)

(−0.6

5)

(−0.2

3)

(0.8

9)

FD

P0.6

20.6

70.6

20.7

30.2

00.6

60.6

9−

0.3

10.9

40.9

70.2

8

(5.6

1)

(5.4

3)

(5.6

5)

(5.9

7)

(1.1

7)

(4.9

4)

(5.3

7)

(−2.6

5)

(7.2

7)

(7.4

5)

(1.6

2)

PO

LI

0.6

2−

6.4

35.3

3−

4.5

5

(0.3

1)

(−2.4

6)

(1.1

2)

(−1.7

3)

OD

M−

2.6

2−

1.9

5

(−4.0

3)

(−3.2

7)

AD

M0.6

6−

0.1

4

(0.7

8)

(−0.1

7)

NO

C35

35

35

35

35

15

15

15

11

11

11

NO

B255

255

255

255

255

155

155

155

68

68

68

AD

J-R

SQ

0.2

70.3

20.2

70.3

10.6

40.2

70.2

70.7

90.5

50.5

40.7

5


4.4. The relationship between the savings rate and the incomeshares of the rich

As mentioned in section 3, many early studies on growth and incomedistribution have assumed a larger savings rate for the rich than for themiddle class and the poor (see Kaldor, 1957, for an explicit mathematicalmodel). If this assumption is valid, an economy with a larger income sharefor the rich should have a higher savings rate, other things given. To testthis classical hypothesis, we consider the following regression

Sit = αi + β1TOP20it + β2GGRWit + β3Rit + β4FDPit + β5POLIit + εit (14)

where TOP20 denotes the income share of the top 20 percent of thepopulation.

Tables 7 to 9 report the estimation results regarding the relationshipbetween the savings rate and the income shares of the rich (TOP20). Re-gardless of whether SR, NSR or NSRP is used as the dependent variable,the coefficients of TOP20 are mostly insignificant, except in cases (4) and(8) in Table 8. When SR is used, TOP20 has negative (insignificant) co-efficients for the complete sample. However, for the two subsamples thecoefficients are all positive. See results in Table 7. As for the results inTables 8 and 9, except for the fixed-effects model, most of the coefficientsare negative, although insignificant. These results seem to weakly suggestthat when the rich’s income share increases, the savings rate drops.

5. CONCLUSION

While this paper has emphasized the analytical ambiguity on the rela-tionship between savings and income inequality, the empirical examinationhas rendered some weak support for a negative association between thesavings rate and income inequality. In particular, with the complete, worldsample, a negative and significant relationship exists between NSR (orNSRP) - the gross national (or private) savings rate - and the Gini coef-ficient. This finding implies that income inequality lowers savings acrosscountries worldwide, and it is in line with the negative relationship betweeninequality and economic growth found by Alesina and Rodrik (1994); andPersson and Tabellini (1994). But this finding is not very robust. In partic-ular, with the subsamples of OECD countries and Asian countries, incomeinequality and the savings rate can even be positively and significantlyassociated.

TA

BLE

7.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Top

20%

ofth

ePopula

tion’s

Inco

me

Share

Dep

enden

tV

aria

ble:

SR

(Gro

ssD

om

esticSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−160.1

8−127.1

4−144.3

9−109.5

3-

−192.5

1−165.0

7-

−229.5

6−217.8

2-

(−6.9

9)

(−4.8

6)

(−6.1

8)

(−4.7

0)

(−6.2

3)

(−6.3

8)

(−2.7

5)

(−3.1

4)

TO

P20

−0.2

5−

0.5

8−

0.1

0−

0.7

1−

0.3

91.0

40.4

40.3

70.3

20.1

90.2

3

(−0.4

9)

(−1.1

1)

(−0.1

9)

(−1.3

9)

(−0.6

9)

(1.3

1)

(0.6

3)

(0.7

9)

(0.2

1)

(0.1

3)

(0.2

1)

GG

RW

1.5

71.7

42.4

71.8

21.6

03.7

03.8

82.1

45.6

45.7

81.3

2

(1.9

3)

(2.1

3)

(3.0

4)

(2.2

4)

(4.5

9)

(4.5

4)

(4.7

9)

(6.6

8)

(2.6

6)

(2.8

3)

(1.5

4)

R−

0.0

3−

0.0

4−

0.0

6−

0.0

3−

0.0

90.6

90.5

9−

0.8

10.6

70.6

40.1

5

(−0.2

8)

(−0.3

5)

(−0.5

2)

(−0.3

4)

(−1.9

5)

(2.0

0)

(1.7

3)

(−5.2

5)

(0.7

7)

(0.7

4)

(0.4

0)

FD

P0.5

20.6

70.4

10.6

3−

0.1

10.1

00.2

0−

0.2

51.1

41.1

1−

0.0

2

(4.4

0)

(5.2

6)

(3.4

4)

(5.0

7)

(−1.1

4)

(0.8

0)

(1.7

0)

(−3.2

5)

(3.9

9)

(4.1

6)

(−0.0

9)

PO

LI

7.6

43.4

97.7

41.3

7

(4.3

9)

(1.4

7)

(1.6

1)

(0.2

6)

OD

M−

2.4

0−

2.9

2

(−3.8

5)

(−5.6

5)

AD

M−

1.4

7−

1.0

7

(−1.6

8)

(−1.2

8)

NO

C37

37

37

37

37

15

15

15

14

14

14

NO

B284

284

284

284

284

168

168

168

86

86

86

AD

J-R

SQ

0.1

30.1

80.0

80.1

70.8

90.1

80.1

70.8

90.2

90.3

00.9

2

TA

BLE

8.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Top

20%

ofth

ePopula

tion’s

Inco

me

Share

Dep

enden

tV

aria

ble:

NSR

(Gro

ssN

atio

nalSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−179.1

3−

153.1

3−153.6

8−120.4

5-

−157.1

1−121.9

6-

−181.8

2−167.0

1-

(−10.2

1)

(−7.6

2)

(−8.2

1)

(−6.7

1)

(−3.9

1)

(−3.6

2)

(−4.0

1)

(−4.4

4)

TO

P20

−0.3

8−

0.6

7−

0.3

1−

0.9

40.4

3−

0.7

1−

1.4

71.7

1−

0.1

3−

0.3

01.4

1

(−1.0

0)

(−1.7

0)

(−0.7

4)

(−2.3

9)

(0.5

8)

(−0.7

0)

(−1.6

2)

(2.0

9)

(−0.1

6)

(−0.3

8)

(1.3

3)

GG

RW

2.7

42.7

03.8

92.8

82.4

05.4

65.7

32.6

23.4

83.6

70.9

7

(4.5

2)

(4.4

1)

(6.1

3)

(4.6

3)

(5.4

9)

(5.0

3)

(5.3

2)

(4.6

0)

(2.9

8)

(3.3

0)

(1.2

1)

R0.0

10.0

2−

0.0

10.0

2−

0.0

8−

0.0

3−

0.1

6−

1.2

60.2

50.2

2−

0.0

5

(0.1

6)

(0.2

1)

(−0.1

7)

(0.2

5)

(−1.3

8)

(−0.0

8)

(−0.3

7)

(−4.8

1)

(0.5

2)

(0.4

6)

(−0.1

5)

FD

P0.7

00.7

70.5

50.7

10.0

50.4

70.5

8−

0.3

50.9

70.9

40.3

5

(7.8

7)

(7.9

8)

(5.8

7)

(7.3

3)

(0.4

4)

(2.8

8)

(3.8

7)

(−2.5

4)

(6.2

7)

(6.4

9)

(2.2

3)

PO

LI

9.2

56.0

79.6

51.7

4

(6.9

8)

(3.3

7)

(1.5

8)

(0.5

9)

OD

M−

1.6

4−

2.5

7

(−3.2

6)

(−6.0

3)

AD

M−

0.4

40.2

5

(−0.6

7)

(0.3

9)

NO

C37

37

37

37

37

15

15

15

14

14

14

NO

B259

259

259

259

259

144

144

144

84

84

84

AD

J-R

SQ

0.3

80.4

00.2

60.3

80.7

70.2

40.2

30.8

20.4

80.4

80.8

3

TA

BLE

9.

Estim

atio

nR

esults:

Rela

tionsh

ipbetw

eenth

eSav

ings

Rate

and

Top

20%

ofth

ePopula

tion’s

Inco

me

Share

Dep

enden

tV

aria

ble:

NSR

P(G

ross

Priva

teSav

ings

Rate)

CO

MP

LET

ESA

MP

LE

OEC

DC

OU

NT

RIE

SA

SIA

NC

OU

NT

RIE

S

IND

.VA

R.

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

CO

NSTA

NT

−164.7

7−154.8

5−157.1

2−133.6

2-

−171.5

1−163.3

5-

−147.0

3−175.8

3-

(−8.7

4)

(−7.5

6)

(−7.9

1)

(−6.9

5)

(−4.3

4)

(−4.9

7)

(−3.8

2)

(−4.6

6)

TO

P20

−0.7

7−

0.7

3−

0.4

3−

0.7

90.4

2−

0.7

8−

0.9

51.0

1−

0.4

6−

0.3

40.8

9

(−1.8

2)

(−1.7

5)

(−0.9

8)

(−1.8

7)

(0.5

0)

(−0.7

8)

(−1.0

8)

(1.0

0)

(−0.6

0)

(−0.4

4)

(0.7

0)

GG

RW

0.7

61.2

31.4

21.1

61.5

13.7

43.8

01.0

21.0

00.4

30.5

1

(1.0

9)

(1.7

4)

(1.9

7)

(1.6

3)

(2.9

7)

(3.5

0)

(3.6

2)

(1.4

5)

(0.9

3)

(0.4

0)

(0.5

3)

R0.1

30.0

90.1

00.0

8−

0.0

50.3

90.3

6−

0.5

2−

0.2

8−

0.0

70.1

9

(1.6

4)

(1.1

2)

(1.1

5)

(1.0

5)

(−0.7

7)

(0.8

9)

(0.8

4)

(−1.6

0)

(−0.6

6)

(−0.1

6)

(0.4

4)

FD

P0.7

20.9

10.6

40.8

60.2

00.8

70.8

9−

0.1

30.9

30.9

80.4

6

(7.8

2)

(8.9

8)

(6.6

4)

(8.5

3)

(1.5

5)

(5.3

5)

(6.0

8)

(−0.7

5)

(6.9

8)

(7.2

0)

(2.4

4)

PO

LI

7.9

35.7

72.2

4−

6.2

6

(5.1

9)

(2.7

3)

(0.3

7)

(−2.3

1)

OD

M−

2.0

4−

2.7

9

(−3.8

7)

(−6.1

4)

AD

M−

2.0

5−

1.3

3

(−2.8

4)

(−1.9

5)

NO

C31

31

31

31

31

15

15

15

11

11

11

NO

B227

227

227

227

227

144

144

144

65

65

65

AD

J-R

SQ

0.3

00.3

50.2

20.3

30.7

10.2

90.2

90.7

30.5

70.5

40.7

2


It is also interesting to note that our empirical estimation on the relation-ship between the income share of the rich and the savings rate does notin general support the classical Kaldor hypothesis: that a larger incomeshare for the rich increases the aggregate savings rate. To some degree, wehave found some evidence for something just the opposite: a larger incomeshare for the rich lowers the private savings rate.

APPENDIX: DATA DESCRIPTION AND SOURCES

Variable Source

(1) Gross domestic savings rate (SR) World Bank, National Account, various issues.

(2) Gross national savings rate (NSR) IMF, World Economic Outlook, various issues.

(3) Domestic private savings rate (NSRP) IMF, World Economic Outlook, various issues.

(4) Gini coefficients (GINI) Deininger and Squire (1996).

(5) Per capita GDP growth (GGRW) World Bank, Social Indicators, various issues.

(6) Real interest rate R Defined as:

R = nominal interest rate − inflation rate

(7) Nominal interest rate IMF, International Financial Statistics,

various issues.

(8) Inflation rate IMF, International Financial Statistics,

various issues.

(9) Financial depth (FINDP) M2 and GDP from: IMF, Government Finance

Statistics and International Financial Statistics,

various issues.

(10) Civil liberty (POLI) Gastil, “Freedom in the World,” various issues;

Barro and Lee, (1994).


REFERENCESAghevli, B., Boughton, J.M., Montiel, P.J., Villanueva, D., and G. Woglom, 1990,The role of national saving in the world economy: Recent trends and prospects. IMFOccasional Paper 67, Washington, D.C., International Monetary Fund.

Alesina, A. and D. Rodrik, 1994, Distributive politics and economic growth. TheQuarterly Journal of Economics 109, 465-490.

Barro, Robert J. and J-W. Lee, 1994, Data Set for a Panel of 138 Countries.

Carroll, C.D., Rhee, B.-K and C. Rhee, 1994, Are there cultural effects on saving?Some cross-sectional evidence. The Quarterly Journal of Economics 109, 685-700.

Deaton, A., 1992, Understanding Consumption, Oxford: Clarendon Press.

Deininger, K., and L. Squire, 1996, A new data set on income distribution. WorldBank Economic Review 10, 565-591.

Della Valle, P. and N. Oguchi, 1976, Distribution, the aggregate consumption func-tion, and the level of economic development: Some cross-country results. Journal ofPolitical Economy 84, 1325-34.

Edwards, S., 1995, Why are saving rates so different across countries?: An interna-tional comparative analysis. NBER working paper #5097.

International Monetary Fund (IMF), World Economic Outlook, various issues.

Kaldor, N., 1957, A model of economic growth. Economic Journal 67, 591-624.

Kessler, D., Perelman, S., and P. Pestieau, 1993, Savings behavior in 17 OECD coun-tries. Review of Income and Wealth 39, 37-49.

Lewis, W.A., 1954, Economic development with unlimited supplies of labor. TheManchester School 22, 139-191.

Li, H., Squire, L., and H. Zou, 1998, Explaining international and intertemporalvariations in income inequality. Economic Journal 108, 26-43.

Li, H., and H. Zou, 1998, Income inequality id not harmful for growth: theory andEvidence. Review of Development Economics 2, 318-334.

Li, H., Xie, D., and H. Zou, 2000, Dynamics of income distribution. Canadian Journalof Economics 33, 937-961.

Li, H., Xu, C. L., and H. Zou, Corruption, income distribution, and growth. Eco-nomics and Politics 12, 155-182.

Lovell, M., 1975, The collective allocation of commodities in a democratic society.Public Choice 24, 71-92.

Maddison, A., 1992, A long-run perspective on saving. The Scandinavian Journal ofEconomics 94, 181-196.

Masson, P.R., Bayoumi, T., and H. Samiei, 1995, Saving behavior in industrial anddeveloping countries, IMF working paper.

Musgrove, P., 1980, Income distribution and the aggregate consumption function.Journal of Political Economy 88, 504-525.

Pasinetti, L., 1962, Rate of profit and income distribution in relation to the rate ofeconomic growth. Review of Economic Studies 29, 267-279.

Perroti, R., 1993, Political equilibrium, income distribution and growth. Review ofEconomic Studies 60, 755-776.

Persson, T. and G. Tabellini, 1994, Is inequality harmful for growth? The AmericanEconomic Review 84, 600-621.


Sahota, G., 1993, Saving and distribution. In: The Economics of Saving. J.H. Gap-inski (ed.) Boston: Kluwer Academic Publishers, pp 193-231.

Shafer, J.R., Elmeskov, J., and W. Tease, 1992, Saving trends and measurementissues. The Scandinavian Journal of Economics 94, 155-175.

World Bank, Social Indicators. various issues.

World Bank, National Accounts. various issues.

savings and income distribution

Documents